KPI-Lens

KPI-Lens

An MCP server for supply chain intelligence that enables conversational querying of eight operational KPIs and anomaly detection data through Claude. It allows users to analyze performance metrics like OTIF and inventory turnover while providing LLM-generated root-cause explanations.

Category
访问服务器

README

KPI-Lens

CI Python License: MIT Release Deploy on Render

An AI-powered supply chain intelligence platform that monitors 8 operational KPIs, detects anomalies using an ensemble of statistical detectors, and explains root causes via Claude — all accessible through a Streamlit dashboard, FastAPI, and an MCP server.

Supply chain teams spend hours manually reviewing KPI dashboards and writing exception reports. KPI-Lens automates the entire loop: ingest → detect → explain → report. A single docker compose up gives you a live anomaly feed, LLM-generated root-cause narratives, and one-click Excel/PPT exports ready for SteerCo.

Features

  • 8 supply chain KPIs tracked weekly: OTIF, Fill Rate, DFA, Inventory Turnover, DIO, Supplier DPPM, Lead Time Variance, PO Cycle Time
  • Ensemble anomaly detection: Z-score + IQR + CUSUM + Isolation Forest detectors with weighted voting
  • LLM root-cause analysis: Claude generates narrative explanations and recommended actions for each anomaly
  • FastAPI backend with 10+ endpoints for KPI data, anomaly management, and LLM chat
  • Streamlit dashboard with 5 pages: Command Center, KPI Deep Dive, Anomaly Log, LLM Analyst, Reports
  • MCP server for Claude Desktop integration — query live KPI data conversationally
  • Automated ingestion: CSV/Excel file watcher with Pydantic v2 validation and APScheduler cron
  • Report generation: Excel workbooks and PowerPoint decks for SteerCo presentations
  • 80%+ test coverage across unit and integration tests; CI runs on Python 3.11 + 3.12

Quick Start

Docker Compose (recommended)

git clone https://github.com/aliivaezii/kpi-lens.git
cd kpi-lens
cp .env.example .env          # Add your ANTHROPIC_API_KEY
docker compose up -d api dashboard

# Seed 2 years of synthetic KPI data (first run only)
docker compose run --rm api python scripts/seed_database.py

# Open the dashboard
open http://localhost:8501

Local Development

python -m venv .venv && source .venv/bin/activate
pip install -e ".[dev]"
cp .env.example .env          # Add your ANTHROPIC_API_KEY

# Seed the database
python data/seeds/generate_kpis.py

# Start services (three terminals)
uvicorn kpi_lens.api.main:app --reload --port 8000
streamlit run kpi_lens/dashboard/app.py
python -m kpi_lens.mcp_server.server   # optional: MCP for Claude Desktop

Architecture

┌─────────────────────────────────────────────────────────────┐
│  External Sources (CSV/Excel exports from ERP)              │
└──────────────────────────┬──────────────────────────────────┘
                           │ ingestion/loader.py + validator.py
                           ▼
┌─────────────────────────────────────────────────────────────┐
│  SQLite DB  ←──  db/repository.py (only DB gateway)        │
└──────┬────────────────────────────────────────────────────┬─┘
       │                                                    │
       ▼                                                    ▼
┌─────────────────────┐                     ┌──────────────────────────┐
│  anomaly/ensemble   │   AnomalyResult     │  api/  (FastAPI)         │
│  ┣ threshold        │ ────────────────►   │  dashboard/ (Streamlit)  │
│  ┣ zscore/iqr/cusum │                     │  mcp_server/ (FastMCP)   │
│  ┗ isolation forest │                     └──────────────────────────┘
└─────────┬───────────┘
          │ async (non-blocking)
          ▼
┌─────────────────────────────────────────────────────────────┐
│  llm/analyst.py  →  Claude via Anthropic SDK               │
│  Generates narrative + recommended actions per anomaly      │
└─────────────────────────────────────────────────────────────┘
          │
          ▼
┌─────────────────────────────────────────────────────────────┐
│  reporting/  →  Excel workbook  +  PowerPoint deck          │
└─────────────────────────────────────────────────────────────┘

KPI Reference

KPI Unit Direction Green threshold Industry Benchmark
OTIF Delivery Rate % Higher is better 95% 95.5%
Order Fill Rate % Higher is better 97% 96%
Demand Forecast Accuracy % Higher is better 85% 80%
Inventory Turnover turns/yr Higher is better 12 10
Days Inventory Outstanding days Lower is better 30 35
Supplier DPPM ppm Lower is better 500 800
Lead Time Variance days Lower is better 3 5
PO Cycle Time days Lower is better 14 18

API Reference

Method Endpoint Description
GET /api/health Health check
GET /api/kpis/snapshot Latest value + health status for all 8 KPIs
GET /api/kpis/{name}/series Time-series data for one KPI
GET /api/kpis/{name}/entities Entity (supplier) breakdown
GET /api/kpis/{name}/benchmarks Industry benchmark percentiles
GET /api/anomalies Recent anomalies with severity filter
POST /api/anomalies/{id}/acknowledge Acknowledge an anomaly
POST /api/llm/chat Chat with the supply chain analyst
POST /api/reports/enqueue Enqueue an anomaly report

Interactive docs: http://localhost:8000/api/docs

Project Structure

kpi_lens/
├── db/           # repository.py — the only DB gateway; schema.py — ORM models
├── kpis/         # definitions.py — 8 KPI constants; snapshot.py — enrichment
├── anomaly/      # base.py, threshold, statistical, ml, ensemble detectors
├── llm/          # client.py (retry), analyst.py, context_builder.py, prompts.py
├── ingestion/    # loader.py, validator.py (Pydantic v2), scheduler.py (APScheduler)
├── reporting/    # excel_exporter.py, powerpoint.py, pdf_converter.py
├── api/          # FastAPI app + routes (kpis, anomalies, llm, reports, health)
├── dashboard/    # Streamlit app + 5 pages
└── mcp_server/   # FastMCP tools for Claude Desktop
config/           # kpis.yaml, anomaly.yaml, report.yaml (change without redeploy)
scripts/          # seed_database.py, run_anomaly_scan.py
tests/
├── unit/         # 8 test files, 70+ tests, no I/O
└── integration/  # FastAPI test client, in-memory DB, mocked LLM

Running Tests

# Unit tests (fast, no infrastructure needed)
pytest tests/unit/ -v --cov=kpi_lens --cov-fail-under=80

# Integration tests (FastAPI + in-memory DB)
pytest tests/integration/ -v

# All tests
pytest tests/ -v --cov=kpi_lens --cov-fail-under=80

Seeding Data

# Default: 104 weeks (2 years) of synthetic data for all 8 KPIs
python scripts/seed_database.py

# Custom parameters
python scripts/seed_database.py --weeks 52

# Run anomaly detection on seeded data
python scripts/run_anomaly_scan.py

Deployment

Render (full stack — API + Dashboard)

Click Deploy on Render above or create a Web Service pointing to this repo. Render reads render.yaml automatically. Set ANTHROPIC_API_KEY in the environment variables panel before deploying.

Streamlit Community Cloud (dashboard only)

  1. Fork this repo
  2. Go to share.streamlit.io → New app
  3. Set Main file path: kpi_lens/dashboard/app.py
  4. Under Advanced settings → Secrets, add:
    ANTHROPIC_API_KEY = "sk-ant-..."
    DATABASE_URL = "sqlite:///kpi_lens.db"
    
  5. Deploy — Streamlit seeds the demo DB on first run

Docker Compose (self-hosted)

git clone https://github.com/aliivaezii/kpi-lens.git
cd kpi-lens
cp .env.example .env   # add ANTHROPIC_API_KEY
docker compose up -d api dashboard
docker compose run --rm api python scripts/seed_database.py
open http://localhost:8501

Data

The platform ships with a realistic synthetic dataset generated by data/seeds/generate_kpis.py:

  • 104 weeks × 8 KPIs × 5 supplier entities = 4,160 weekly records
  • Additive model with seasonality, trend, and deliberate anomaly injection at known dates
  • Designed to exercise all four detector types (threshold, Z-score, IQR, CUSUM)

To use your own data, drop a CSV into data/imports/ and run the ingestion scheduler, or POST directly to POST /api/ingest. Format reference: data/samples/sample_kpi_data.csv.

License

MIT — see LICENSE.

推荐服务器

Baidu Map

Baidu Map

百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。

官方
精选
JavaScript
Playwright MCP Server

Playwright MCP Server

一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。

官方
精选
TypeScript
Magic Component Platform (MCP)

Magic Component Platform (MCP)

一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。

官方
精选
本地
TypeScript
Audiense Insights MCP Server

Audiense Insights MCP Server

通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。

官方
精选
本地
TypeScript
VeyraX

VeyraX

一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。

官方
精选
本地
graphlit-mcp-server

graphlit-mcp-server

模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。

官方
精选
TypeScript
Kagi MCP Server

Kagi MCP Server

一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。

官方
精选
Python
e2b-mcp-server

e2b-mcp-server

使用 MCP 通过 e2b 运行代码。

官方
精选
Neon MCP Server

Neon MCP Server

用于与 Neon 管理 API 和数据库交互的 MCP 服务器

官方
精选
Exa MCP Server

Exa MCP Server

模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。

官方
精选